Establishing detection thresholds for environmental DNA using receiver operator characteristic (ROC) curves

Abstract

Environmental DNA (eDNA) detection is increasingly being used to assess the potential presence of aquatic species. Setting defensible thresholds for qPCR assays are necessary to differentiate “detection” from “non-detection”, but threshold rationales are rarely reported in eDNA studies. Detection thresholds for data inclusion and exclusion have implications for downstream data analysis; uncritical acceptance could result in false positive results (Type I error), whereas overly conservative thresholds could result in high false negative interpretation (Type II error). We assessed the use of the receiver operator characteristics (ROC) framework to identify a detection threshold for balancing sensitivity and specificity of eDNA data. The ROC framework was tested using empirical eDNA data from redside dace (Clinostomus elongatus), an endangered freshwater fish in southwestern Ontario. Based on negative controls, positive controls, and quality assurance guidelines, a minimum concentration of 1 copy reaction−1 was identified as the most appropriate threshold for this study. Although optimal thresholds will likely vary between studies and labs based on empirical data, the ROC framework provides a statistical method for establishing defensible thresholds for data acceptance or rejection as well as tradeoffs between the sensitivity and specificity of eDNA data.